[PATCH 0/4] sched/fair: Burstable CFS bandwidth controller

[PATCH 0/4] sched/fair: Burstable CFS bandwidth controller

[Huaixin Chang]

This patchset is rebased upon v5.11-rc6.
Previous archive:
https://lore.kernel.org/lkml/20201217074620.58338-1-changhuaixin@linux.alibaba.com/

The CFS bandwidth controller limits CPU requests of a task group to
quota during each period. However, parallel workloads might be bursty
so that they get throttled. And they are latency sensitive at the same
time so that throttling them is undesired.

Scaling up period and quota allows greater burst capacity. But it might
cause longer stuck till next refill. We introduce "burst" to allow
accumulating unused quota from previous periods, and to be assigned when
a task group requests more CPU than quota during a specific period. Thus
allowing CPU time requests as long as the average requested CPU time is
below quota on the long run. The maximum accumulation is capped by burst
and is set 0 by default, thus the traditional behaviour remains.

A huge drop of 99th tail latency from more than 500ms to 27ms is seen for
real java workloads when using burst. Similar drops are seen when
testing with schbench too:

	echo $$ > /sys/fs/cgroup/cpu/test/cgroup.procs
	echo 700000 > /sys/fs/cgroup/cpu/test/cpu.cfs_quota_us
	echo 100000 > /sys/fs/cgroup/cpu/test/cpu.cfs_period_us
	echo 400000 > /sys/fs/cgroup/cpu/test/cpu.cfs_burst_us

	# The average CPU usage is around 500%, which is 200ms CPU time
	# every 40ms.
	./schbench -m 1 -t 30 -r 60 -c 10000 -R 500

	Without burst:

	Latency percentiles (usec)
	50.0000th: 7
	75.0000th: 8
	90.0000th: 9
	95.0000th: 10
	*99.0000th: 933
	99.5000th: 981
	99.9000th: 3068
	min=0, max=20054
	rps: 498.31 p95 (usec) 10 p99 (usec) 933 p95/cputime 0.10% p99/cputime 9.33%

	With burst:

	Latency percentiles (usec)
	50.0000th: 7
	75.0000th: 8
	90.0000th: 9
	95.0000th: 9
	*99.0000th: 12
	99.5000th: 13
	99.9000th: 19
	min=0, max=406
	rps: 498.36 p95 (usec) 9 p99 (usec) 12 p95/cputime 0.09% p99/cputime 0.12%

How much workloads with benefit from burstable CFS bandwidth control
depends on how bursty and how latency sensitive they are.

Previously, Cong Wang and Konstantin Khlebnikov proposed similar
feature:
https://lore.kernel.org/lkml/20180522062017.5193-1-xiyou.wangcong@gmail.com/
https://lore.kernel.org/lkml/157476581065.5793.4518979877345136813.stgit@buzz/

This time we present more latency statistics and handle overflow while
accumulating.

Huaixin Chang (4):
  sched/fair: Introduce primitives for CFS bandwidth burst
  sched/fair: Make CFS bandwidth controller burstable
  sched/fair: Add cfs bandwidth burst statistics
  sched/fair: Add document for burstable CFS bandwidth control

 Documentation/scheduler/sched-bwc.rst |  49 +++++++++++--
 include/linux/sched/sysctl.h          |   2 +
 kernel/sched/core.c                   | 126 +++++++++++++++++++++++++++++-----
 kernel/sched/fair.c                   |  58 +++++++++++++---
 kernel/sched/sched.h                  |   9 ++-
 kernel/sysctl.c                       |  18 +++++
 6 files changed, 232 insertions(+), 30 deletions(-)

-- 
2.14.4.44.g2045bb6

[PATCH 1/4] sched/fair: Introduce primitives for CFS bandwidth burst

[Huaixin Chang]

In this patch, we introduce the notion of CFS bandwidth burst. Unused
"quota" from pervious "periods" might be accumulated and used in the
following "periods". The maximum amount of accumulated bandwidth is
bounded by "burst". And the maximun amount of CPU a group can consume in
a given period is "buffer" which is equivalent to "quota" + "burst in
case that this group has done enough accumulation.

Signed-off-by: Huaixin Chang <changhuaixin@linux.alibaba.com>
Signed-off-by: Shanpei Chen <shanpeic@linux.alibaba.com>
---
 kernel/sched/core.c  | 91 ++++++++++++++++++++++++++++++++++++++++++++--------
 kernel/sched/fair.c  |  2 ++
 kernel/sched/sched.h |  2 ++
 3 files changed, 82 insertions(+), 13 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index ff74fca39ed2..28e3165c685b 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -8590,10 +8590,12 @@ static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC;
 
 static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);
 
-static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
+static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota,
+							u64 burst)
 {
 	int i, ret = 0, runtime_enabled, runtime_was_enabled;
 	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
+	u64 buffer;
 
 	if (tg == &root_task_group)
 		return -EINVAL;
@@ -8620,6 +8622,16 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
 	if (quota != RUNTIME_INF && quota > max_cfs_runtime)
 		return -EINVAL;
 
+	/*
+	 * Bound burst to defend burst against overflow during bandwidth shift.
+	 */
+	if (burst > max_cfs_runtime)
+		return -EINVAL;
+
+	if (quota == RUNTIME_INF)
+		buffer = RUNTIME_INF;
+	else
+		buffer = min(max_cfs_runtime, quota + burst);
 	/*
 	 * Prevent race between setting of cfs_rq->runtime_enabled and
 	 * unthrottle_offline_cfs_rqs().
@@ -8641,6 +8653,8 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
 	raw_spin_lock_irq(&cfs_b->lock);
 	cfs_b->period = ns_to_ktime(period);
 	cfs_b->quota = quota;
+	cfs_b->burst = burst;
+	cfs_b->buffer = buffer;
 
 	__refill_cfs_bandwidth_runtime(cfs_b);
 
@@ -8674,9 +8688,10 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
 
 static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
 {
-	u64 quota, period;
+	u64 quota, period, burst;
 
 	period = ktime_to_ns(tg->cfs_bandwidth.period);
+	burst = tg->cfs_bandwidth.burst;
 	if (cfs_quota_us < 0)
 		quota = RUNTIME_INF;
 	else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC)
@@ -8684,7 +8699,7 @@ static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
 	else
 		return -EINVAL;
 
-	return tg_set_cfs_bandwidth(tg, period, quota);
+	return tg_set_cfs_bandwidth(tg, period, quota, burst);
 }
 
 static long tg_get_cfs_quota(struct task_group *tg)
@@ -8702,15 +8717,16 @@ static long tg_get_cfs_quota(struct task_group *tg)
 
 static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us)
 {
-	u64 quota, period;
+	u64 quota, period, burst;
 
 	if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC)
 		return -EINVAL;
 
 	period = (u64)cfs_period_us * NSEC_PER_USEC;
 	quota = tg->cfs_bandwidth.quota;
+	burst = tg->cfs_bandwidth.burst;
 
-	return tg_set_cfs_bandwidth(tg, period, quota);
+	return tg_set_cfs_bandwidth(tg, period, quota, burst);
 }
 
 static long tg_get_cfs_period(struct task_group *tg)
@@ -8723,6 +8739,35 @@ static long tg_get_cfs_period(struct task_group *tg)
 	return cfs_period_us;
 }
 
+static int tg_set_cfs_burst(struct task_group *tg, long cfs_burst_us)
+{
+	u64 quota, period, burst;
+
+	period = ktime_to_ns(tg->cfs_bandwidth.period);
+	quota = tg->cfs_bandwidth.quota;
+	if (cfs_burst_us < 0)
+		burst = RUNTIME_INF;
+	else if ((u64)cfs_burst_us <= U64_MAX / NSEC_PER_USEC)
+		burst = (u64)cfs_burst_us * NSEC_PER_USEC;
+	else
+		return -EINVAL;
+
+	return tg_set_cfs_bandwidth(tg, period, quota, burst);
+}
+
+static long tg_get_cfs_burst(struct task_group *tg)
+{
+	u64 burst_us;
+
+	if (tg->cfs_bandwidth.burst == RUNTIME_INF)
+		return -1;
+
+	burst_us = tg->cfs_bandwidth.burst;
+	do_div(burst_us, NSEC_PER_USEC);
+
+	return burst_us;
+}
+
 static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
 				  struct cftype *cft)
 {
@@ -8747,6 +8792,18 @@ static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
 	return tg_set_cfs_period(css_tg(css), cfs_period_us);
 }
 
+static s64 cpu_cfs_burst_read_s64(struct cgroup_subsys_state *css,
+				  struct cftype *cft)
+{
+	return tg_get_cfs_burst(css_tg(css));
+}
+
+static int cpu_cfs_burst_write_s64(struct cgroup_subsys_state *css,
+				   struct cftype *cftype, s64 cfs_burst_us)
+{
+	return tg_set_cfs_burst(css_tg(css), cfs_burst_us);
+}
+
 struct cfs_schedulable_data {
 	struct task_group *tg;
 	u64 period, quota;
@@ -8899,6 +8956,11 @@ static struct cftype cpu_legacy_files[] = {
 		.read_u64 = cpu_cfs_period_read_u64,
 		.write_u64 = cpu_cfs_period_write_u64,
 	},
+	{
+		.name = "cfs_burst_us",
+		.read_s64 = cpu_cfs_burst_read_s64,
+		.write_s64 = cpu_cfs_burst_write_s64,
+	},
 	{
 		.name = "stat",
 		.seq_show = cpu_cfs_stat_show,
@@ -9019,26 +9081,27 @@ static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css,
 #endif
 
 static void __maybe_unused cpu_period_quota_print(struct seq_file *sf,
-						  long period, long quota)
+					long period, long quota, long burst)
 {
 	if (quota < 0)
 		seq_puts(sf, "max");
 	else
 		seq_printf(sf, "%ld", quota);
 
-	seq_printf(sf, " %ld\n", period);
+	seq_printf(sf, " %ld %ld\n", period, burst);
 }
 
-/* caller should put the current value in *@periodp before calling */
+/* caller should put the current value in *@periodp and *@burstp before calling */
 static int __maybe_unused cpu_period_quota_parse(char *buf,
-						 u64 *periodp, u64 *quotap)
+					 u64 *periodp, u64 *quotap, u64 *burstp)
 {
 	char tok[21];	/* U64_MAX */
 
-	if (sscanf(buf, "%20s %llu", tok, periodp) < 1)
+	if (sscanf(buf, "%20s %llu %llu", tok, periodp, burstp) < 1)
 		return -EINVAL;
 
 	*periodp *= NSEC_PER_USEC;
+	*burstp *= NSEC_PER_USEC;
 
 	if (sscanf(tok, "%llu", quotap))
 		*quotap *= NSEC_PER_USEC;
@@ -9055,7 +9118,8 @@ static int cpu_max_show(struct seq_file *sf, void *v)
 {
 	struct task_group *tg = css_tg(seq_css(sf));
 
-	cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg));
+	cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg),
+			tg_get_cfs_burst(tg));
 	return 0;
 }
 
@@ -9064,12 +9128,13 @@ static ssize_t cpu_max_write(struct kernfs_open_file *of,
 {
 	struct task_group *tg = css_tg(of_css(of));
 	u64 period = tg_get_cfs_period(tg);
+	u64 burst = tg_get_cfs_burst(tg);
 	u64 quota;
 	int ret;
 
-	ret = cpu_period_quota_parse(buf, &period, &quota);
+	ret = cpu_period_quota_parse(buf, &period, &quota, &burst);
 	if (!ret)
-		ret = tg_set_cfs_bandwidth(tg, period, quota);
+		ret = tg_set_cfs_bandwidth(tg, period, quota, burst);
 	return ret ?: nbytes;
 }
 #endif
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 04a3ce20da67..46945349f209 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -5262,6 +5262,8 @@ void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
 	cfs_b->runtime = 0;
 	cfs_b->quota = RUNTIME_INF;
 	cfs_b->period = ns_to_ktime(default_cfs_period());
+	cfs_b->burst = 0;
+	cfs_b->buffer = RUNTIME_INF;
 
 	INIT_LIST_HEAD(&cfs_b->throttled_cfs_rq);
 	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index bb09988451a0..2c0d8469c0fb 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -357,6 +357,8 @@ struct cfs_bandwidth {
 	ktime_t			period;
 	u64			quota;
 	u64			runtime;
+	u64			burst;
+	u64			buffer;
 	s64			hierarchical_quota;
 
 	u8			idle;
-- 
2.14.4.44.g2045bb6

[PATCH 2/4] sched/fair: Make CFS bandwidth controller burstable

[Huaixin Chang]

Accumulate unused quota from previous periods, thus accumulated
bandwidth runtime can be used in the following periods. During
accumulation, take care of runtime overflow. Previous non-burstable
CFS bandwidth controller only assign quota to runtime, that saves a lot.

A sysctl parameter sysctl_sched_cfs_bw_burst_onset_percent is introduced to
denote how many percent of burst is given on setting cfs bandwidth. By
default it is 0, which means on burst is allowed unless accumulated.

Also, parameter sysctl_sched_cfs_bw_burst_enabled is introduced as a
switch for burst. It is enabled by default.

Signed-off-by: Huaixin Chang <changhuaixin@linux.alibaba.com>
Signed-off-by: Shanpei Chen <shanpeic@linux.alibaba.com>
Reported-by: kernel test robot <lkp@intel.com>
---
 include/linux/sched/sysctl.h |  2 ++
 kernel/sched/core.c          | 31 +++++++++++++++++++++++++----
 kernel/sched/fair.c          | 47 ++++++++++++++++++++++++++++++++++++--------
 kernel/sched/sched.h         |  4 ++--
 kernel/sysctl.c              | 18 +++++++++++++++++
 5 files changed, 88 insertions(+), 14 deletions(-)

diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h
index 3c31ba88aca5..3400828eaf2d 100644
--- a/include/linux/sched/sysctl.h
+++ b/include/linux/sched/sysctl.h
@@ -72,6 +72,8 @@ extern unsigned int sysctl_sched_uclamp_util_min_rt_default;
 
 #ifdef CONFIG_CFS_BANDWIDTH
 extern unsigned int sysctl_sched_cfs_bandwidth_slice;
+extern unsigned int sysctl_sched_cfs_bw_burst_onset_percent;
+extern unsigned int sysctl_sched_cfs_bw_burst_enabled;
 #endif
 
 #ifdef CONFIG_SCHED_AUTOGROUP
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 28e3165c685b..9f1b05ad0411 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -66,6 +66,16 @@ const_debug unsigned int sysctl_sched_features =
  */
 const_debug unsigned int sysctl_sched_nr_migrate = 32;
 
+#ifdef CONFIG_CFS_BANDWIDTH
+/*
+ * Percent of burst assigned to cfs_b->runtime on tg_set_cfs_bandwidth,
+ * 0 by default.
+ */
+unsigned int sysctl_sched_cfs_bw_burst_onset_percent;
+
+unsigned int sysctl_sched_cfs_bw_burst_enabled = 1;
+#endif
+
 /*
  * period over which we measure -rt task CPU usage in us.
  * default: 1s
@@ -8586,7 +8596,7 @@ static DEFINE_MUTEX(cfs_constraints_mutex);
 const u64 max_cfs_quota_period = 1 * NSEC_PER_SEC; /* 1s */
 static const u64 min_cfs_quota_period = 1 * NSEC_PER_MSEC; /* 1ms */
 /* More than 203 days if BW_SHIFT equals 20. */
-static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC;
+const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC;
 
 static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);
 
@@ -8595,7 +8605,7 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota,
 {
 	int i, ret = 0, runtime_enabled, runtime_was_enabled;
 	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
-	u64 buffer;
+	u64 buffer, burst_onset;
 
 	if (tg == &root_task_group)
 		return -EINVAL;
@@ -8656,11 +8666,24 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota,
 	cfs_b->burst = burst;
 	cfs_b->buffer = buffer;
 
-	__refill_cfs_bandwidth_runtime(cfs_b);
+	cfs_b->max_overrun = DIV_ROUND_UP_ULL(max_cfs_runtime, quota);
+	cfs_b->runtime = cfs_b->quota;
+
+	/* burst_onset needed */
+	if (cfs_b->quota != RUNTIME_INF &&
+			sysctl_sched_cfs_bw_burst_enabled &&
+			sysctl_sched_cfs_bw_burst_onset_percent > 0) {
+
+		burst_onset = div_u64(burst, 100) *
+			sysctl_sched_cfs_bw_burst_onset_percent;
+
+		cfs_b->runtime += burst_onset;
+		cfs_b->runtime = min(max_cfs_runtime, cfs_b->runtime);
+	}
 
 	/* Restart the period timer (if active) to handle new period expiry: */
 	if (runtime_enabled)
-		start_cfs_bandwidth(cfs_b);
+		start_cfs_bandwidth(cfs_b, 1);
 
 	raw_spin_unlock_irq(&cfs_b->lock);
 
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 46945349f209..6a7c261d206a 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4609,10 +4609,23 @@ static inline u64 sched_cfs_bandwidth_slice(void)
  *
  * requires cfs_b->lock
  */
-void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b)
+static void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b,
+		u64 overrun)
 {
-	if (cfs_b->quota != RUNTIME_INF)
-		cfs_b->runtime = cfs_b->quota;
+	u64 refill;
+
+	if (cfs_b->quota != RUNTIME_INF) {
+
+		if (!sysctl_sched_cfs_bw_burst_enabled) {
+			cfs_b->runtime = cfs_b->quota;
+			return;
+		}
+
+		overrun = min(overrun, cfs_b->max_overrun);
+		refill = cfs_b->quota * overrun;
+		cfs_b->runtime += refill;
+		cfs_b->runtime = min(cfs_b->runtime, cfs_b->buffer);
+	}
 }
 
 static inline struct cfs_bandwidth *tg_cfs_bandwidth(struct task_group *tg)
@@ -4634,7 +4647,7 @@ static int __assign_cfs_rq_runtime(struct cfs_bandwidth *cfs_b,
 	if (cfs_b->quota == RUNTIME_INF)
 		amount = min_amount;
 	else {
-		start_cfs_bandwidth(cfs_b);
+		start_cfs_bandwidth(cfs_b, 0);
 
 		if (cfs_b->runtime > 0) {
 			amount = min(cfs_b->runtime, min_amount);
@@ -4980,7 +4993,7 @@ static int do_sched_cfs_period_timer(struct cfs_bandwidth *cfs_b, int overrun, u
 	if (cfs_b->idle && !throttled)
 		goto out_deactivate;
 
-	__refill_cfs_bandwidth_runtime(cfs_b);
+	__refill_cfs_bandwidth_runtime(cfs_b, overrun);
 
 	if (!throttled) {
 		/* mark as potentially idle for the upcoming period */
@@ -5201,6 +5214,7 @@ static enum hrtimer_restart sched_cfs_slack_timer(struct hrtimer *timer)
 }
 
 extern const u64 max_cfs_quota_period;
+extern const u64 max_cfs_runtime;
 
 static enum hrtimer_restart sched_cfs_period_timer(struct hrtimer *timer)
 {
@@ -5230,7 +5244,14 @@ static enum hrtimer_restart sched_cfs_period_timer(struct hrtimer *timer)
 			new = old * 2;
 			if (new < max_cfs_quota_period) {
 				cfs_b->period = ns_to_ktime(new);
-				cfs_b->quota *= 2;
+				cfs_b->quota = min(cfs_b->quota * 2,
+						max_cfs_runtime);
+
+				cfs_b->buffer = min(max_cfs_runtime,
+						cfs_b->quota + cfs_b->burst);
+				/* Add 1 in case max_overrun becomes 0. */
+				cfs_b->max_overrun >>= 1;
+				cfs_b->max_overrun++;
 
 				pr_warn_ratelimited(
 	"cfs_period_timer[cpu%d]: period too short, scaling up (new cfs_period_us = %lld, cfs_quota_us = %lld)\n",
@@ -5279,16 +5300,26 @@ static void init_cfs_rq_runtime(struct cfs_rq *cfs_rq)
 	INIT_LIST_HEAD(&cfs_rq->throttled_list);
 }
 
-void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
+void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b, int init)
 {
+	u64 overrun;
+
 	lockdep_assert_held(&cfs_b->lock);
 
 	if (cfs_b->period_active)
 		return;
 
 	cfs_b->period_active = 1;
-	hrtimer_forward_now(&cfs_b->period_timer, cfs_b->period);
+	overrun = hrtimer_forward_now(&cfs_b->period_timer, cfs_b->period);
 	hrtimer_start_expires(&cfs_b->period_timer, HRTIMER_MODE_ABS_PINNED);
+
+	/*
+	 * When period timer stops, quota for the following period is not
+	 * refilled, however period timer is already forwarded. We should
+	 * accumulate quota once more than overrun here.
+	 */
+	if (!init)
+		__refill_cfs_bandwidth_runtime(cfs_b, overrun + 1);
 }
 
 static void destroy_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 2c0d8469c0fb..4f53ea8e92ce 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -359,6 +359,7 @@ struct cfs_bandwidth {
 	u64			runtime;
 	u64			burst;
 	u64			buffer;
+	u64			max_overrun;
 	s64			hierarchical_quota;
 
 	u8			idle;
@@ -469,8 +470,7 @@ extern void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq,
 			struct sched_entity *parent);
 extern void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
 
-extern void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b);
-extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b);
+extern void start_cfs_bandwidth(struct cfs_bandwidth *cfs_b, int init);
 extern void unthrottle_cfs_rq(struct cfs_rq *cfs_rq);
 
 extern void free_rt_sched_group(struct task_group *tg);
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index c9fbdd848138..6839a37ee29c 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -1842,6 +1842,24 @@ static struct ctl_table kern_table[] = {
 		.proc_handler	= proc_dointvec_minmax,
 		.extra1		= SYSCTL_ONE,
 	},
+	{
+		.procname	= "sched_cfs_bw_burst_onset_percent",
+		.data		= &sysctl_sched_cfs_bw_burst_onset_percent,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= &one_hundred,
+	},
+	{
+		.procname	= "sched_cfs_bw_burst_enabled",
+		.data		= &sysctl_sched_cfs_bw_burst_enabled,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE,
+	},
 #endif
 #if defined(CONFIG_ENERGY_MODEL) && defined(CONFIG_CPU_FREQ_GOV_SCHEDUTIL)
 	{
-- 
2.14.4.44.g2045bb6

[PATCH 3/4] sched/fair: Add cfs bandwidth burst statistics

[Huaixin Chang]

Introduce statistics exports for the burstable cfs bandwidth
controller.

The following exports are included:

current_bw: current runtime in global pool
nr_burst:   number of periods bandwidth burst occurs
burst_time: cumulative wall-time that any cpus has
	    used above quota in respective periods

Signed-off-by: Huaixin Chang <changhuaixin@linux.alibaba.com>
Signed-off-by: Shanpei Chen <shanpeic@linux.alibaba.com>
---
 kernel/sched/core.c  |  6 ++++++
 kernel/sched/fair.c  | 12 +++++++++++-
 kernel/sched/sched.h |  3 +++
 3 files changed, 20 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 9f1b05ad0411..d253903dbb4e 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -8681,6 +8681,8 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota,
 		cfs_b->runtime = min(max_cfs_runtime, cfs_b->runtime);
 	}
 
+	cfs_b->previous_runtime = cfs_b->runtime;
+
 	/* Restart the period timer (if active) to handle new period expiry: */
 	if (runtime_enabled)
 		start_cfs_bandwidth(cfs_b, 1);
@@ -8929,6 +8931,10 @@ static int cpu_cfs_stat_show(struct seq_file *sf, void *v)
 		seq_printf(sf, "wait_sum %llu\n", ws);
 	}
 
+	seq_printf(sf, "current_bw %llu\n", cfs_b->runtime);
+	seq_printf(sf, "nr_burst %d\n", cfs_b->nr_burst);
+	seq_printf(sf, "burst_time %llu\n", cfs_b->burst_time);
+
 	return 0;
 }
 #endif /* CONFIG_CFS_BANDWIDTH */
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 6a7c261d206a..4cd3dc16659c 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4612,7 +4612,7 @@ static inline u64 sched_cfs_bandwidth_slice(void)
 static void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b,
 		u64 overrun)
 {
-	u64 refill;
+	u64 refill, runtime;
 
 	if (cfs_b->quota != RUNTIME_INF) {
 
@@ -4621,10 +4621,20 @@ static void __refill_cfs_bandwidth_runtime(struct cfs_bandwidth *cfs_b,
 			return;
 		}
 
+		if (cfs_b->previous_runtime > cfs_b->runtime) {
+			runtime = cfs_b->previous_runtime - cfs_b->runtime;
+			if (runtime > cfs_b->quota) {
+				cfs_b->burst_time += runtime - cfs_b->quota;
+				cfs_b->nr_burst++;
+			}
+		}
+
 		overrun = min(overrun, cfs_b->max_overrun);
 		refill = cfs_b->quota * overrun;
 		cfs_b->runtime += refill;
 		cfs_b->runtime = min(cfs_b->runtime, cfs_b->buffer);
+
+		cfs_b->previous_runtime = cfs_b->runtime;
 	}
 }
 
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 4f53ea8e92ce..04b0a1ce4c89 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -360,6 +360,7 @@ struct cfs_bandwidth {
 	u64			burst;
 	u64			buffer;
 	u64			max_overrun;
+	u64			previous_runtime;
 	s64			hierarchical_quota;
 
 	u8			idle;
@@ -372,7 +373,9 @@ struct cfs_bandwidth {
 	/* Statistics: */
 	int			nr_periods;
 	int			nr_throttled;
+	int			nr_burst;
 	u64			throttled_time;
+	u64			burst_time;
 #endif
 };
 
-- 
2.14.4.44.g2045bb6

[PATCH 4/4] sched/fair: Add document for burstable CFS bandwidth control

[Huaixin Chang]

Basic description of usage and effect for CFS Bandwidth Control Burst.

Signed-off-by: Huaixin Chang <changhuaixin@linux.alibaba.com>
Signed-off-by: Shanpei Chen <shanpeic@linux.alibaba.com>
---
 Documentation/scheduler/sched-bwc.rst | 70 +++++++++++++++++++++++++++++++++--
 1 file changed, 66 insertions(+), 4 deletions(-)

diff --git a/Documentation/scheduler/sched-bwc.rst b/Documentation/scheduler/sched-bwc.rst
index 9801d6b284b1..0933c66cc68b 100644
--- a/Documentation/scheduler/sched-bwc.rst
+++ b/Documentation/scheduler/sched-bwc.rst
@@ -21,18 +21,46 @@ cfs_quota units at each period boundary. As threads consume this bandwidth it
 is transferred to cpu-local "silos" on a demand basis. The amount transferred
 within each of these updates is tunable and described as the "slice".
 
+By default, CPU bandwidth consumption is strictly limited to quota within each
+given period. For the sequence of CPU usage u_i served under CFS bandwidth
+control, if for any j <= k N(j,k) is the number of periods from u_j to u_k:
+
+        u_j+...+u_k <= quota * N(j,k)
+
+For a bursty sequence among which interval u_j...u_k are at the peak, CPU
+requests might have to wait for more periods to replenish enough quota.
+Otherwise, larger quota is required.
+
+With "burst" buffer, CPU requests might be served as long as:
+
+        u_j+...+u_k <= B_j + quota * N(j,k)
+
+if for any j <= k N(j,k) is the number of periods from u_j to u_k and B_j is
+the accumulated quota from previous periods in burst buffer serving u_j.
+Burst buffer helps in that serving whole bursty CPU requests without throttling
+them can be done with moderate quota setting and accumulated quota in burst
+buffer, if:
+
+        u_0+...+u_n <= B_0 + quota * N(0,n)
+
+where B_0 is the initial state of burst buffer. The maximum accumulated quota in
+the burst buffer is capped by burst. With proper burst setting, the available
+bandwidth is still determined by quota and period on the long run.
+
 Management
 ----------
-Quota and period are managed within the cpu subsystem via cgroupfs.
+Quota, period and burst are managed within the cpu subsystem via cgroupfs.
 
-cpu.cfs_quota_us: the total available run-time within a period (in microseconds)
+cpu.cfs_quota_us: run-time replenished within a period (in microseconds)
 cpu.cfs_period_us: the length of a period (in microseconds)
+cpu.cfs_burst_us: the maximum accumulated run-time (in microseconds)
 cpu.stat: exports throttling statistics [explained further below]
 
 The default values are::
 
 	cpu.cfs_period_us=100ms
-	cpu.cfs_quota=-1
+	cpu.cfs_quota_us=-1
+	cpu.cfs_burst_us=0
 
 A value of -1 for cpu.cfs_quota_us indicates that the group does not have any
 bandwidth restriction in place, such a group is described as an unconstrained
@@ -48,6 +76,11 @@ more detail below.
 Writing any negative value to cpu.cfs_quota_us will remove the bandwidth limit
 and return the group to an unconstrained state once more.
 
+A value of 0 for cpu.cfs_burst_us indicates that the group can not accumulate
+any unused bandwidth. It makes the traditional bandwidth control behavior for
+CFS unchanged. Writing any (valid) positive value(s) into cpu.cfs_burst_us
+will enact the cap on unused bandwidth accumulation.
+
 Any updates to a group's bandwidth specification will result in it becoming
 unthrottled if it is in a constrained state.
 
@@ -65,9 +98,21 @@ This is tunable via procfs::
 Larger slice values will reduce transfer overheads, while smaller values allow
 for more fine-grained consumption.
 
+There is also a global switch to turn off burst for all groups::
+       /proc/sys/kernel/sched_cfs_bw_burst_enabled (default=1)
+
+By default it is enabled. Writing a 0 value means no accumulated CPU time can be
+used for any group, even if cpu.cfs_burst_us is configured.
+
+Sometimes users might want a group to burst without accumulation. This is
+tunable via::
+       /proc/sys/kernel/sched_cfs_bw_burst_onset_percent (default=0)
+
+Up to 100% runtime of cpu.cfs_burst_us might be given on setting bandwidth.
+
 Statistics
 ----------
-A group's bandwidth statistics are exported via 3 fields in cpu.stat.
+A group's bandwidth statistics are exported via 6 fields in cpu.stat.
 
 cpu.stat:
 
@@ -75,6 +120,11 @@ cpu.stat:
 - nr_throttled: Number of times the group has been throttled/limited.
 - throttled_time: The total time duration (in nanoseconds) for which entities
   of the group have been throttled.
+- current_bw: Current runtime in global pool.
+- nr_burst: Number of periods burst occurs.
+- burst_time: Cumulative wall-time that any CPUs has used above quota in
+  respective periods
+
 
 This interface is read-only.
 
@@ -172,3 +222,15 @@ Examples
 
    By using a small period here we are ensuring a consistent latency
    response at the expense of burst capacity.
+
+4. Limit a group to 20% of 1 CPU, and allow accumulate up to 60% of 1 CPU
+   additionally, in case accumulation has been done.
+
+   With 50ms period, 10ms quota will be equivalent to 20% of 1 CPU.
+   And 30ms burst will be equivalent to 60% of 1 CPU.
+
+	# echo 10000 > cpu.cfs_quota_us /* quota = 10ms */
+	# echo 50000 > cpu.cfs_period_us /* period = 50ms */
+	# echo 30000 > cpu.cfs_burst_us /* burst = 30ms */
+
+   Larger buffer setting allows greater burst capacity.
-- 
2.14.4.44.g2045bb6

[PATCH 1/4] sched/fair: Introduce primitives for CFS bandwidth burst

[Huaixin Chang]

In this patch, we introduce the notion of CFS bandwidth burst. Unused
"quota" from pervious "periods" might be accumulated and used in the
following "periods". The maximum amount of accumulated bandwidth is
bounded by "burst". And the maximun amount of CPU a group can consume in
a given period is "buffer" which is equivalent to "quota" + "burst in
case that this group has done enough accumulation.

Signed-off-by: Huaixin Chang <changhuaixin@linux.alibaba.com>
Signed-off-by: Shanpei Chen <shanpeic@linux.alibaba.com>
---
 kernel/sched/core.c  | 91 ++++++++++++++++++++++++++++++++++++++++++++--------
 kernel/sched/fair.c  |  2 ++
 kernel/sched/sched.h |  2 ++
 3 files changed, 82 insertions(+), 13 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index e7e453492cff..48d3bad12be2 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -7895,10 +7895,12 @@ static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC;
 
 static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);
 
-static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
+static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota,
+							u64 burst)
 {
 	int i, ret = 0, runtime_enabled, runtime_was_enabled;
 	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
+	u64 buffer;
 
 	if (tg == &root_task_group)
 		return -EINVAL;
@@ -7925,6 +7927,16 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
 	if (quota != RUNTIME_INF && quota > max_cfs_runtime)
 		return -EINVAL;
 
+	/*
+	 * Bound burst to defend burst against overflow during bandwidth shift.
+	 */
+	if (burst > max_cfs_runtime)
+		return -EINVAL;
+
+	if (quota == RUNTIME_INF)
+		buffer = RUNTIME_INF;
+	else
+		buffer = min(max_cfs_runtime, quota + burst);
 	/*
 	 * Prevent race between setting of cfs_rq->runtime_enabled and
 	 * unthrottle_offline_cfs_rqs().
@@ -7946,6 +7958,8 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
 	raw_spin_lock_irq(&cfs_b->lock);
 	cfs_b->period = ns_to_ktime(period);
 	cfs_b->quota = quota;
+	cfs_b->burst = burst;
+	cfs_b->buffer = buffer;
 
 	__refill_cfs_bandwidth_runtime(cfs_b);
 
@@ -7979,9 +7993,10 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
 
 static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
 {
-	u64 quota, period;
+	u64 quota, period, burst;
 
 	period = ktime_to_ns(tg->cfs_bandwidth.period);
+	burst = tg->cfs_bandwidth.burst;
 	if (cfs_quota_us < 0)
 		quota = RUNTIME_INF;
 	else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC)
@@ -7989,7 +8004,7 @@ static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
 	else
 		return -EINVAL;
 
-	return tg_set_cfs_bandwidth(tg, period, quota);
+	return tg_set_cfs_bandwidth(tg, period, quota, burst);
 }
 
 static long tg_get_cfs_quota(struct task_group *tg)
@@ -8007,15 +8022,16 @@ static long tg_get_cfs_quota(struct task_group *tg)
 
 static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us)
 {
-	u64 quota, period;
+	u64 quota, period, burst;
 
 	if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC)
 		return -EINVAL;
 
 	period = (u64)cfs_period_us * NSEC_PER_USEC;
 	quota = tg->cfs_bandwidth.quota;
+	burst = tg->cfs_bandwidth.burst;
 
-	return tg_set_cfs_bandwidth(tg, period, quota);
+	return tg_set_cfs_bandwidth(tg, period, quota, burst);
 }
 
 static long tg_get_cfs_period(struct task_group *tg)
@@ -8028,6 +8044,35 @@ static long tg_get_cfs_period(struct task_group *tg)
 	return cfs_period_us;
 }
 
+static int tg_set_cfs_burst(struct task_group *tg, long cfs_burst_us)
+{
+	u64 quota, period, burst;
+
+	period = ktime_to_ns(tg->cfs_bandwidth.period);
+	quota = tg->cfs_bandwidth.quota;
+	if (cfs_burst_us < 0)
+		burst = RUNTIME_INF;
+	else if ((u64)cfs_burst_us <= U64_MAX / NSEC_PER_USEC)
+		burst = (u64)cfs_burst_us * NSEC_PER_USEC;
+	else
+		return -EINVAL;
+
+	return tg_set_cfs_bandwidth(tg, period, quota, burst);
+}
+
+static long tg_get_cfs_burst(struct task_group *tg)
+{
+	u64 burst_us;
+
+	if (tg->cfs_bandwidth.burst == RUNTIME_INF)
+		return -1;
+
+	burst_us = tg->cfs_bandwidth.burst;
+	do_div(burst_us, NSEC_PER_USEC);
+
+	return burst_us;
+}
+
 static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
 				  struct cftype *cft)
 {
@@ -8052,6 +8097,18 @@ static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
 	return tg_set_cfs_period(css_tg(css), cfs_period_us);
 }
 
+static s64 cpu_cfs_burst_read_s64(struct cgroup_subsys_state *css,
+				  struct cftype *cft)
+{
+	return tg_get_cfs_burst(css_tg(css));
+}
+
+static int cpu_cfs_burst_write_s64(struct cgroup_subsys_state *css,
+				   struct cftype *cftype, s64 cfs_burst_us)
+{
+	return tg_set_cfs_burst(css_tg(css), cfs_burst_us);
+}
+
 struct cfs_schedulable_data {
 	struct task_group *tg;
 	u64 period, quota;
@@ -8204,6 +8261,11 @@ static struct cftype cpu_legacy_files[] = {
 		.read_u64 = cpu_cfs_period_read_u64,
 		.write_u64 = cpu_cfs_period_write_u64,
 	},
+	{
+		.name = "cfs_burst_us",
+		.read_s64 = cpu_cfs_burst_read_s64,
+		.write_s64 = cpu_cfs_burst_write_s64,
+	},
 	{
 		.name = "stat",
 		.seq_show = cpu_cfs_stat_show,
@@ -8324,26 +8386,27 @@ static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css,
 #endif
 
 static void __maybe_unused cpu_period_quota_print(struct seq_file *sf,
-						  long period, long quota)
+					long period, long quota, long burst)
 {
 	if (quota < 0)
 		seq_puts(sf, "max");
 	else
 		seq_printf(sf, "%ld", quota);
 
-	seq_printf(sf, " %ld\n", period);
+	seq_printf(sf, " %ld %ld\n", period, burst);
 }
 
-/* caller should put the current value in *@periodp before calling */
+/* caller should put the current value in *@periodp and *@burstp before calling */
 static int __maybe_unused cpu_period_quota_parse(char *buf,
-						 u64 *periodp, u64 *quotap)
+					 u64 *periodp, u64 *quotap, u64 *burstp)
 {
 	char tok[21];	/* U64_MAX */
 
-	if (sscanf(buf, "%20s %llu", tok, periodp) < 1)
+	if (sscanf(buf, "%20s %llu %llu", tok, periodp, burstp) < 1)
 		return -EINVAL;
 
 	*periodp *= NSEC_PER_USEC;
+	*burstp *= NSEC_PER_USEC;
 
 	if (sscanf(tok, "%llu", quotap))
 		*quotap *= NSEC_PER_USEC;
@@ -8360,7 +8423,8 @@ static int cpu_max_show(struct seq_file *sf, void *v)
 {
 	struct task_group *tg = css_tg(seq_css(sf));
 
-	cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg));
+	cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg),
+			tg_get_cfs_burst(tg));
 	return 0;
 }
 
@@ -8369,12 +8433,13 @@ static ssize_t cpu_max_write(struct kernfs_open_file *of,
 {
 	struct task_group *tg = css_tg(of_css(of));
 	u64 period = tg_get_cfs_period(tg);
+	u64 burst = tg_get_cfs_burst(tg);
 	u64 quota;
 	int ret;
 
-	ret = cpu_period_quota_parse(buf, &period, &quota);
+	ret = cpu_period_quota_parse(buf, &period, &quota, &burst);
 	if (!ret)
-		ret = tg_set_cfs_bandwidth(tg, period, quota);
+		ret = tg_set_cfs_bandwidth(tg, period, quota, burst);
 	return ret ?: nbytes;
 }
 #endif
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index ae7ceba8fd4f..6bb4f89259fd 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -5242,6 +5242,8 @@ void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
 	cfs_b->runtime = 0;
 	cfs_b->quota = RUNTIME_INF;
 	cfs_b->period = ns_to_ktime(default_cfs_period());
+	cfs_b->burst = 0;
+	cfs_b->buffer = RUNTIME_INF;
 
 	INIT_LIST_HEAD(&cfs_b->throttled_cfs_rq);
 	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index df80bfcea92e..a8772eca8cbb 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -364,6 +364,8 @@ struct cfs_bandwidth {
 	ktime_t			period;
 	u64			quota;
 	u64			runtime;
+	u64			burst;
+	u64			buffer;
 	s64			hierarchical_quota;
 
 	u8			idle;
-- 
2.14.4.44.g2045bb6

Re: [PATCH 1/4] sched/fair: Introduce primitives for CFS bandwidth burst

[changhuaixin]

> On Dec 21, 2020, at 9:53 PM, changhuaixin <changhuaixin@linux.alibaba.com> wrote:
> 
> 
> 
>> On Dec 17, 2020, at 9:36 PM, Peter Zijlstra <peterz@infradead.org> wrote:
>> 
>> On Thu, Dec 17, 2020 at 03:46:17PM +0800, Huaixin Chang wrote:
>>> In this patch, we introduce the notion of CFS bandwidth burst. Unused
>>> "quota" from pervious "periods" might be accumulated and used in the
>>> following "periods". The maximum amount of accumulated bandwidth is
>>> bounded by "burst". And the maximun amount of CPU a group can consume in
>>> a given period is "buffer" which is equivalent to "quota" + "burst in
>>> case that this group has done enough accumulation.
>> 
>> Oh man, Juri, wasn't there a paper about statistical bandwidth
>> accounting somewhere? Where, if you replace every utilization by a
>> statistical variable, the end result is still useful?
>> 
>> That is, instead of something like; \Sum u_i <= 1, you get something
>> like: \Sum {avg(u),var(u)}_i <= {1, sqrt(\Sum var_i^2)} and you can
>> still proof bounded tardiness etc.. (assuming a gaussian distribution).
>> 
>> The proposed seems close to that, but not quite, and I'm afraid it's not
>> quite strong enough to still provide any guarantees.
> 
> After reading some papers on statistical bandwidth sharing, it occurs to me that statistical bandwidth sharing is about the way bandwidth is shared between competitors. I wonder if the paper you mentioned was "Insensitivity results in statistical bandwidth sharing" or some paper referenced, which showed the end result is insensitive to maybe the distribution or the arrival pattern.
> 
> I am sorry that I cannot prove using statistical bandwidth sharing theory now. However, I wonder if it is more acceptable to put rate-based cfsb after share fairness, because the input stream of cfsb account is the output stream of fairness. And that is in contrast with what statistical bandwidth sharing does, in which fairness is used to share bandwidth upon the output stream of rate-based control. In this way, maybe guarantees can be provided by share fairness, and cfsb may use a larger buffer to allow more jitters.
> 
> A buffer, which is several times of quota, is able to handle large bursts, and throttle threads soon when overloaded. The present cfsb, however, has to be configured several times larger to handle jitters, which is ineffective and does not provide the designed rate-based control.

Hi, peter.

Will you please have a look at my earlier reply.

The guarantee provided by cfsb is max-min fairness with strict upper bound, I think. And after this modification, it is still roughly that. As for longer periods, each cfsb group is still limited to their quota on average.

When users configure a huge burst buffer, cfsb doesn't really throttle threads and upper bound is exceeded. In that case, guarantees are provided by share, which provides proportional fairness.

Thanks,
Huaixin

Re: [PATCH 1/4] sched/fair: Introduce primitives for CFS bandwidth burst

[changhuaixin]

> On Dec 17, 2020, at 9:36 PM, Peter Zijlstra <peterz@infradead.org> wrote:
> 
> On Thu, Dec 17, 2020 at 03:46:17PM +0800, Huaixin Chang wrote:
>> In this patch, we introduce the notion of CFS bandwidth burst. Unused
>> "quota" from pervious "periods" might be accumulated and used in the
>> following "periods". The maximum amount of accumulated bandwidth is
>> bounded by "burst". And the maximun amount of CPU a group can consume in
>> a given period is "buffer" which is equivalent to "quota" + "burst in
>> case that this group has done enough accumulation.
> 
> Oh man, Juri, wasn't there a paper about statistical bandwidth
> accounting somewhere? Where, if you replace every utilization by a
> statistical variable, the end result is still useful?
> 
> That is, instead of something like; \Sum u_i <= 1, you get something
> like: \Sum {avg(u),var(u)}_i <= {1, sqrt(\Sum var_i^2)} and you can
> still proof bounded tardiness etc.. (assuming a gaussian distribution).
> 
> The proposed seems close to that, but not quite, and I'm afraid it's not
> quite strong enough to still provide any guarantees.

After reading some papers on statistical bandwidth sharing, it occurs to me that statistical bandwidth sharing is about the way bandwidth is shared between competitors. I wonder if the paper you mentioned was "Insensitivity results in statistical bandwidth sharing" or some paper referenced, which showed the end result is insensitive to maybe the distribution or the arrival pattern.

I am sorry that I cannot prove using statistical bandwidth sharing theory now. However, I wonder if it is more acceptable to put rate-based cfsb after share fairness, because the input stream of cfsb account is the output stream of fairness. And that is in contrast with what statistical bandwidth sharing does, in which fairness is used to share bandwidth upon the output stream of rate-based control. In this way, maybe guarantees can be provided by share fairness, and cfsb may use a larger buffer to allow more jitters.

A buffer, which is several times of quota, is able to handle large bursts, and throttle threads soon when overloaded. The present cfsb, however, has to be configured several times larger to handle jitters, which is ineffective and does not provide the designed rate-based control.

Re: [PATCH 1/4] sched/fair: Introduce primitives for CFS bandwidth burst

[Peter Zijlstra]

On Thu, Dec 17, 2020 at 03:46:17PM +0800, Huaixin Chang wrote:
> In this patch, we introduce the notion of CFS bandwidth burst. Unused
> "quota" from pervious "periods" might be accumulated and used in the
> following "periods". The maximum amount of accumulated bandwidth is
> bounded by "burst". And the maximun amount of CPU a group can consume in
> a given period is "buffer" which is equivalent to "quota" + "burst in
> case that this group has done enough accumulation.

Oh man, Juri, wasn't there a paper about statistical bandwidth
accounting somewhere? Where, if you replace every utilization by a
statistical variable, the end result is still useful?

That is, instead of something like; \Sum u_i <= 1, you get something
like: \Sum {avg(u),var(u)}_i <= {1, sqrt(\Sum var_i^2)} and you can
still proof bounded tardiness etc.. (assuming a gaussian distribution).

The proposed seems close to that, but not quite, and I'm afraid it's not
quite strong enough to still provide any guarantees.

[PATCH 1/4] sched/fair: Introduce primitives for CFS bandwidth burst

[Huaixin Chang]

In this patch, we introduce the notion of CFS bandwidth burst. Unused
"quota" from pervious "periods" might be accumulated and used in the
following "periods". The maximum amount of accumulated bandwidth is
bounded by "burst". And the maximun amount of CPU a group can consume in
a given period is "buffer" which is equivalent to "quota" + "burst in
case that this group has done enough accumulation.

Signed-off-by: Huaixin Chang <changhuaixin@linux.alibaba.com>
Signed-off-by: Shanpei Chen <shanpeic@linux.alibaba.com>
---
 kernel/sched/core.c  | 91 ++++++++++++++++++++++++++++++++++++++++++++--------
 kernel/sched/fair.c  |  2 ++
 kernel/sched/sched.h |  2 ++
 3 files changed, 82 insertions(+), 13 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index e7e453492cff..48d3bad12be2 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -7895,10 +7895,12 @@ static const u64 max_cfs_runtime = MAX_BW * NSEC_PER_USEC;
 
 static int __cfs_schedulable(struct task_group *tg, u64 period, u64 runtime);
 
-static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
+static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota,
+							u64 burst)
 {
 	int i, ret = 0, runtime_enabled, runtime_was_enabled;
 	struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
+	u64 buffer;
 
 	if (tg == &root_task_group)
 		return -EINVAL;
@@ -7925,6 +7927,16 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
 	if (quota != RUNTIME_INF && quota > max_cfs_runtime)
 		return -EINVAL;
 
+	/*
+	 * Bound burst to defend burst against overflow during bandwidth shift.
+	 */
+	if (burst > max_cfs_runtime)
+		return -EINVAL;
+
+	if (quota == RUNTIME_INF)
+		buffer = RUNTIME_INF;
+	else
+		buffer = min(max_cfs_runtime, quota + burst);
 	/*
 	 * Prevent race between setting of cfs_rq->runtime_enabled and
 	 * unthrottle_offline_cfs_rqs().
@@ -7946,6 +7958,8 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
 	raw_spin_lock_irq(&cfs_b->lock);
 	cfs_b->period = ns_to_ktime(period);
 	cfs_b->quota = quota;
+	cfs_b->burst = burst;
+	cfs_b->buffer = buffer;
 
 	__refill_cfs_bandwidth_runtime(cfs_b);
 
@@ -7979,9 +7993,10 @@ static int tg_set_cfs_bandwidth(struct task_group *tg, u64 period, u64 quota)
 
 static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
 {
-	u64 quota, period;
+	u64 quota, period, burst;
 
 	period = ktime_to_ns(tg->cfs_bandwidth.period);
+	burst = tg->cfs_bandwidth.burst;
 	if (cfs_quota_us < 0)
 		quota = RUNTIME_INF;
 	else if ((u64)cfs_quota_us <= U64_MAX / NSEC_PER_USEC)
@@ -7989,7 +8004,7 @@ static int tg_set_cfs_quota(struct task_group *tg, long cfs_quota_us)
 	else
 		return -EINVAL;
 
-	return tg_set_cfs_bandwidth(tg, period, quota);
+	return tg_set_cfs_bandwidth(tg, period, quota, burst);
 }
 
 static long tg_get_cfs_quota(struct task_group *tg)
@@ -8007,15 +8022,16 @@ static long tg_get_cfs_quota(struct task_group *tg)
 
 static int tg_set_cfs_period(struct task_group *tg, long cfs_period_us)
 {
-	u64 quota, period;
+	u64 quota, period, burst;
 
 	if ((u64)cfs_period_us > U64_MAX / NSEC_PER_USEC)
 		return -EINVAL;
 
 	period = (u64)cfs_period_us * NSEC_PER_USEC;
 	quota = tg->cfs_bandwidth.quota;
+	burst = tg->cfs_bandwidth.burst;
 
-	return tg_set_cfs_bandwidth(tg, period, quota);
+	return tg_set_cfs_bandwidth(tg, period, quota, burst);
 }
 
 static long tg_get_cfs_period(struct task_group *tg)
@@ -8028,6 +8044,35 @@ static long tg_get_cfs_period(struct task_group *tg)
 	return cfs_period_us;
 }
 
+static int tg_set_cfs_burst(struct task_group *tg, long cfs_burst_us)
+{
+	u64 quota, period, burst;
+
+	period = ktime_to_ns(tg->cfs_bandwidth.period);
+	quota = tg->cfs_bandwidth.quota;
+	if (cfs_burst_us < 0)
+		burst = RUNTIME_INF;
+	else if ((u64)cfs_burst_us <= U64_MAX / NSEC_PER_USEC)
+		burst = (u64)cfs_burst_us * NSEC_PER_USEC;
+	else
+		return -EINVAL;
+
+	return tg_set_cfs_bandwidth(tg, period, quota, burst);
+}
+
+static long tg_get_cfs_burst(struct task_group *tg)
+{
+	u64 burst_us;
+
+	if (tg->cfs_bandwidth.burst == RUNTIME_INF)
+		return -1;
+
+	burst_us = tg->cfs_bandwidth.burst;
+	do_div(burst_us, NSEC_PER_USEC);
+
+	return burst_us;
+}
+
 static s64 cpu_cfs_quota_read_s64(struct cgroup_subsys_state *css,
 				  struct cftype *cft)
 {
@@ -8052,6 +8097,18 @@ static int cpu_cfs_period_write_u64(struct cgroup_subsys_state *css,
 	return tg_set_cfs_period(css_tg(css), cfs_period_us);
 }
 
+static s64 cpu_cfs_burst_read_s64(struct cgroup_subsys_state *css,
+				  struct cftype *cft)
+{
+	return tg_get_cfs_burst(css_tg(css));
+}
+
+static int cpu_cfs_burst_write_s64(struct cgroup_subsys_state *css,
+				   struct cftype *cftype, s64 cfs_burst_us)
+{
+	return tg_set_cfs_burst(css_tg(css), cfs_burst_us);
+}
+
 struct cfs_schedulable_data {
 	struct task_group *tg;
 	u64 period, quota;
@@ -8204,6 +8261,11 @@ static struct cftype cpu_legacy_files[] = {
 		.read_u64 = cpu_cfs_period_read_u64,
 		.write_u64 = cpu_cfs_period_write_u64,
 	},
+	{
+		.name = "cfs_burst_us",
+		.read_s64 = cpu_cfs_burst_read_s64,
+		.write_s64 = cpu_cfs_burst_write_s64,
+	},
 	{
 		.name = "stat",
 		.seq_show = cpu_cfs_stat_show,
@@ -8324,26 +8386,27 @@ static int cpu_weight_nice_write_s64(struct cgroup_subsys_state *css,
 #endif
 
 static void __maybe_unused cpu_period_quota_print(struct seq_file *sf,
-						  long period, long quota)
+					long period, long quota, long burst)
 {
 	if (quota < 0)
 		seq_puts(sf, "max");
 	else
 		seq_printf(sf, "%ld", quota);
 
-	seq_printf(sf, " %ld\n", period);
+	seq_printf(sf, " %ld %ld\n", period, burst);
 }
 
-/* caller should put the current value in *@periodp before calling */
+/* caller should put the current value in *@periodp and *@burstp before calling */
 static int __maybe_unused cpu_period_quota_parse(char *buf,
-						 u64 *periodp, u64 *quotap)
+					 u64 *periodp, u64 *quotap, u64 *burstp)
 {
 	char tok[21];	/* U64_MAX */
 
-	if (sscanf(buf, "%20s %llu", tok, periodp) < 1)
+	if (sscanf(buf, "%20s %llu %llu", tok, periodp, burstp) < 1)
 		return -EINVAL;
 
 	*periodp *= NSEC_PER_USEC;
+	*burstp *= NSEC_PER_USEC;
 
 	if (sscanf(tok, "%llu", quotap))
 		*quotap *= NSEC_PER_USEC;
@@ -8360,7 +8423,8 @@ static int cpu_max_show(struct seq_file *sf, void *v)
 {
 	struct task_group *tg = css_tg(seq_css(sf));
 
-	cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg));
+	cpu_period_quota_print(sf, tg_get_cfs_period(tg), tg_get_cfs_quota(tg),
+			tg_get_cfs_burst(tg));
 	return 0;
 }
 
@@ -8369,12 +8433,13 @@ static ssize_t cpu_max_write(struct kernfs_open_file *of,
 {
 	struct task_group *tg = css_tg(of_css(of));
 	u64 period = tg_get_cfs_period(tg);
+	u64 burst = tg_get_cfs_burst(tg);
 	u64 quota;
 	int ret;
 
-	ret = cpu_period_quota_parse(buf, &period, &quota);
+	ret = cpu_period_quota_parse(buf, &period, &quota, &burst);
 	if (!ret)
-		ret = tg_set_cfs_bandwidth(tg, period, quota);
+		ret = tg_set_cfs_bandwidth(tg, period, quota, burst);
 	return ret ?: nbytes;
 }
 #endif
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index ae7ceba8fd4f..6bb4f89259fd 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -5242,6 +5242,8 @@ void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
 	cfs_b->runtime = 0;
 	cfs_b->quota = RUNTIME_INF;
 	cfs_b->period = ns_to_ktime(default_cfs_period());
+	cfs_b->burst = 0;
+	cfs_b->buffer = RUNTIME_INF;
 
 	INIT_LIST_HEAD(&cfs_b->throttled_cfs_rq);
 	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index df80bfcea92e..a8772eca8cbb 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -364,6 +364,8 @@ struct cfs_bandwidth {
 	ktime_t			period;
 	u64			quota;
 	u64			runtime;
+	u64			burst;
+	u64			buffer;
 	s64			hierarchical_quota;
 
 	u8			idle;
-- 
2.14.4.44.g2045bb6